AI Worms: Future Threat or Hype?

Anthropic's research suggests small AI models could achieve self-replication within months, forming autonomous botnets. While AI companies frequently highlight potential dangers, experts generally agree on the near-term possibility of such replication. However, the actual threat posed may be less novel than it appears, drawing parallels to past internet worms and botnets.

Experts like Michael Bailey, who has studied internet worms since the early 2000s, note that past botnets such as Mirai already demonstrated the ability to compromise heterogeneous devices. Bailey suggests that while AI could automate the adaptation process to unfamiliar devices, the core functionality of botnets remains similar. He argues that the "self-replicating" framing might be a misnomer, likening it more to a malware dropper that installs coordination tools.

Concerns are raised about whether such research is partly for marketing, especially given Anthropic's upcoming IPO. However, some believe companies are genuinely worried about competitors' recklessness. Proposed mitigation strategies, such as ISP notifications and limiting internet access, are described as "old school." Regulation, like the UK's PSTI Act banning default passwords, is also suggested.

Ultimately, experts argue that defending against potential AI worms will rely on established cybersecurity practices. These include diligent device patching, network isolation, strong passwords, and overall good digital hygiene. The adaptability of AI models, combined with the persistent failure to implement basic security measures, presents an ongoing challenge.